Web intelligence for measuring emerging economic trends: the drone industry - PIET DAAS, MARTIJN TENNEKES, BLANCA DE MÍGUEL, MARÍA DE MIGUEL ...

Page created by Zachary Estrada
 
CONTINUE READING
Web intelligence for measuring emerging economic trends: the drone industry - PIET DAAS, MARTIJN TENNEKES, BLANCA DE MÍGUEL, MARÍA DE MIGUEL ...
Web intelligence for measuring
 emerging economic trends:
 the drone industry
 PIET DAAS, MARTIJN TENNEKES,
 BLANCA DE MÍGUEL, MARÍA DE MIGUEL,
 VIRGINIA SANTAMARINA, FLORABELA CARAUSU 2022 edition

 S TAT I S T I C A L
 W O R K I N G PA P E R S
Web intelligence for measuring emerging economic trends: the drone industry - PIET DAAS, MARTIJN TENNEKES, BLANCA DE MÍGUEL, MARÍA DE MIGUEL ...
Web intelligence for measuring
 emerging economic trends:
 the drone industry 2022 edition

 mixed-mode surveys
Manuscript completed in June 2022

This document should not be considered as representative of the European Commission’s official
position.

Luxembourg: Publications Office of the European Union, 2022

© European Union, 2022

The reuse policy of European Commission documents is implemented based on Commission
Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents (OJ L 330,
14.12.2011, p. 39). Except otherwise noted, the reuse of this document is authorised under a Creative
Commons Attribution 4.0 International (CC-BY 4.0) licence
(https://creativecommons.org/licenses/by/4.0/). This means that reuse is allowed provided appropriate
credit is given and any changes are indicated.

For any use or reproduction of elements that are not owned by the European Union, permission may
need to be sought directly from the respective rightholders. The European Union does not own the
copyright in relation to the following elements:
For more information, please consult: https://ec.europa.eu/eurostat/about/policies/copyright

Copyright for the photograph: Cover © mtp26/Shutterstock

The information and views set out in this publication are those of the authors and do not necessarily
reflect the official opinion of the European Union. Neither the European Union institutions and bodies
nor any person acting on their behalf may be held responsible for the use which may be made of the
information contained therein.

Theme: General and regional statistics
Collection: Statistical working paper

ISBN 978-92-76-53307-8 ISSN 2315-0807 doi: 10.2785/620917 KS-TC-22-004-EN-N
Abstract

Abstract
In order to reap the benefits of the data revolution and to complement traditional statistical methods
and sources with innovative approaches, Eurostat recently established the Trusted Smart Statistics
initiative. This also encompasses the so-called Web Intelligence Hub. In the context of the Web
Intelligence Hub, Eurostat initiated experimental research towards extending the capabilities of
retrieving information on European drone businesses from the internet.
The ‘Web Intelligence for Drones’ initiative builds on previous research into web scraping of business
information in the context of official statistics. The study was triggered by the absence of
consolidated data on drones, the operation of drones or market size in the EU, despite the EU having
one of the most advanced regulations on unmanned aircrafts in the world. In particular, research
focusses on the development of a methodology and of the tools needed to retrieve information from
the World Wide Web (www) concerning businesses based in EU countries that have their main
activity in the civil drones sector.
This scientific summary presents the chain-like methodology and the tools developed to identify
drone companies through the www and to extract company-relevant information from their websites.
The method was developed with a perspective of generalisation in mind (across countries and
across economic sectors) wherever possible. It has already been fully applied to three European
countries (Spain, Italy and Ireland).

Keywords: Drones, Web Intelligence, BigData, new data sources, business statistics
Authors: Piet Daas(1), Martijn Tennekes(2), Blanca de Míguel(3), María de Miguel(4),Virginia
Santamarina(5), Florabela Carausu(6)

Acknowledgments: We would particularly like to thank the coordinators of this study at Eurostat –
Unit A5: Andrea Ascheri, Fernando Reis, Albrecht Wirthmann and Matyas Meszaros for their helpful
comments.
Acknowledgments are extended as well to Paola Olivares (Osservatori Digital Innovation
School of Management – Politecnico di Milano) for her support with the research and for her sharing
sector knowledge about the Drone market in Italy.

The study was implemented by GOPA as a contractor of Eurostat for the framework contract on
Methodological Support (Ref. 2018.0086) in collaboration with Statistics Netherlands together with
individual researchers of the Universitat Politècnica de Valencia.

(1) pjh.daas@cbs.nl
(2) m.tennekes@cbs.nl
(3) bdemigu@upvnet.upv.es
(4) mademi@omp.upv.es
(5) virsanca@upv.es
(6) florabela.carausu@gopa.de

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 3
Table of contents

Table of contents
 Introduction ........................................................................................................... 9

 1. Background ................................................................................................. 10
 1.1. Previous ESS research initiatives on web scraping of business information ............. 10
 1.2. Legal background: the Drones Regulation in Europe ................................................. 12
 1.3. Analysing the industry through company websites ..................................................... 12
 1.4. Exploratory analyses of the drone sector ................................................................... 15

 2. Approaches to searching for company websites ..................................... 18
 2.1. Searching for websites that provide an overview of drone companies ....................... 19
 2.2. Searching for websites of individual drone companies ............................................... 19
 2.3. Finding drone websites via pictures of drones ........................................................... 19
 2.4. Finding drone websites via a company’s impressum page ........................................ 20
 2.5. Finding drone websites using a non-search engine approach ................................... 20

 3. URL retrieval ................................................................................................ 22
 3.1. URL composition explained ........................................................................................ 22
 3.2. URL collection methodology ....................................................................................... 23
 3.2.1. Search queries......................................................................................................... 23
 3.2.2. Search engines ........................................................................................................ 24
 3.2.3. Sequence of steps ................................................................................................... 27
 3.3. Results ........................................................................................................................ 31

 4. Classification of drone company websites ............................................... 34
 4.1. Drone website classification approaches ................................................................... 34
 Classification approaches considered ................................................................................. 35
 4.1.1. Word-based approach ............................................................................................. 35
 4.1.2. PUlearning approach ............................................................................................... 36
 4.1.3. Supervised Machine Learning approach ................................................................. 37
 4.2. Generalising the model ............................................................................................... 39
 4.2.1. The Spanish model .................................................................................................. 39
 4.2.2. Creating a (more) generic model ............................................................................. 40
 4.2.2.1. Applying the model to Irish websites .................................................................. 40
 4.2.2.2. Applying the model to Italian websites ............................................................... 42

 5. Data extraction ............................................................................................ 46
 5.1. Information extracted from potential Drone websites ................................................. 46
 5.1.1. Name of the website owner ..................................................................................... 47
 5.1.2. Short description of the website............................................................................... 47

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 4
Table of contents

 5.1.3. Contact address....................................................................................................... 47
 5.1.4. Region of location .................................................................................................... 48
 5.1.5. Email address .......................................................................................................... 48
 5.1.6. Telephone numbers ................................................................................................. 49
 5.1.7. VAT number ............................................................................................................. 49
 5.1.8. Activities reported .................................................................................................... 49
 5.1.9. Social media presence ............................................................................................ 50
 5.1.10. E-commerce activity ............................................................................................ 50
 5.1.11. Job advertisement presence ............................................................................... 50
 5.1.12. website Start date ............................................................................................... 50
 5.1.13. Is the website really about drones? .................................................................... 50
 5.1.14. Is the website’s owner active in the country studied? ......................................... 51
 5.1.15. Is it a company website? ..................................................................................... 51
 5.1.16. Is it the website of a drone company active in the country? ............................... 52
 5.1.17. Comparison of findings ....................................................................................... 52
 5.2. Results ........................................................................................................................ 53

6. Generalising the methodology and tools .................................................. 59
 6.1. Main considerations .................................................................................................... 59
 6.2. Generic keyword extraction ........................................................................................ 60
 6.2.1. Method ..................................................................................................................... 60
 6.2.2. Results ..................................................................................................................... 61
 6.3. Discussion .................................................................................................................. 65

7. Conclusions and recommendations .......................................................... 66

References: ......................................................................................................... 69

Annex I: Creating a list of drone company websites for Spain ........................ 72

Annex II: Hard- and software requirements ...................................................... 73

Annex III: Top 20 features in the Logistic Regression model .......................... 75

Annex IV: URLs of companies found for Spain ................................................ 76

Annex V: URLs of companies found for Italy .................................................... 78

Annex VI: URLs of companies found for Ireland .............................................. 80

Annex VII: Generic keyword extraction: implementing the method ................ 81

Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 5
Tables

 Table of contents

Tables
Table 1: Main characteristics of previous studies exploring company websites as a source of
information for industry analyses ....................................................................................................... 14
Table 2: Overlap between the secondary and top-domain names of the URLs found by the six search
engines for all Script 1 queries (Version 2) for Spain ......................................................................... 25
Table 3: Overview of scripts developed: input, output, script deployed and run-time ........................ 28
Table 4: Number of links found by the final version of each script, for each Member State studied and
for each language .............................................................................................................................. 31
Table 5: Manual classification results for the drone-specific word-classification findings .................. 36
Table 6: Manual classification results for the trained PUlearning based model ................................. 37
Table 7: Manual classification results for the trained Logistic Regression model .............................. 39
Table 8: Manual classification results for the trained Logistic Regression model at various probability
ranges ................................................................................................................................................ 39
Table 9: Manual classification results for the classification of Irish websites ..................................... 41
Table 10: Manual classification results for the classification of Irish websites at various probability
ranges ................................................................................................................................................ 42
Table 11: Manual classification results for the classification of Italian websites ................................ 44
Table 12: Manual classification results for the classification of Italian websites at various probability
ranges ................................................................................................................................................ 45
Table 13: Percentage of results obtained for the variables extracted for Spain, Ireland, Italy and
countries combined ............................................................................................................................ 52
Table 14: Results of the web search strategy and data extractions for Italy, Spain and Ireland ........ 53
Table 15: Drone company activities in the value chain and their co-occurrence in Spain ................. 54
Table 16: Drone company activities in the value chain and their co-occurrence in Italy .................... 54
Table 17: Drone company activities in the value chain and their co-occurrence in Ireland ............... 55
Table 18: Output table for the Wikipedia page ‘Unmanned Aerial Vehicle’. Only the top 20 rows are
shown. ................................................................................................................................................ 62
Table 19: Output table for the Wikipedia page ‘Circular economy’. Only the top 20 rows are shown. 63
Table 20: Output table for the Wikipedia page ‘Renewable energy’. Only the top 20 rows are shown.
........................................................................................................................................................... 64

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 6
Figures

 Table of contents

Figures
Figure 1: Upset plot of the overlap between the secondary and top-domain names found by the six
search engines for the English and Spanish results combined (Spain Script version 2) ................... 26
Figure 2: Overview of the scripts developed to find Drone companies in various countries .............. 33
Figure 3: Histogram of the model probabilities of being a drone website for Spanish websites ........ 38
Figure 4: Histogram of the model probabilities of being a drone website for Irish websites .............. 41
Figure 5: Histogram of the model probabilities of being a drone website for Italian websites ............ 44
Figure 6: Location of drone companies in Spain ................................................................................ 56
Figure 7: Location of drone companies in Italy .................................................................................. 57
Figure 8: Location of Drone companies in Ireland ............................................................................. 58

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 7
Abbreviations

 Table of contents

Abbreviations
AESA Agencia Estatal de Seguridad Aérea (Spain)
API Application Programming Interface
CBS Centraal Bureau voor de Statistiek (the Netherlands)
D&B Dun & Bradstreet
DTM Document Team Matrix
EASA European Union Aviation Safety Agency
EC European Commission
ENAV Ente Nazionale Assistenza al Volo (Italy)
ESS European Statistical System
ESSnet European Statistical System research network
EU European Union
FAA Federal Aviation Administration
GDP Gross Domestic Product
GDPR General Data Protection Regulation
GISCO Geographic Information System for the Commission
GOPA Gesellschaft für Organisation, Planung und Ausbildung
IAA Irish Aviation Authority
IP Internet Protocol
NACE Statistical classification of economic activities in the European Community
NLTK Natural Language Toolkit
NUTS Nomenclature of Territorial Units for Statistics
RPAS Remotely Piloted Aircraft System
SESAR Single European Sky ATM Research
SMEs Small and Medium Enterprises
UAAI Unmanned Aircraft Association of Ireland
UAS Unmanned Aircraft System
UAV Unmanned Aircraft Vehicle
URL Uniform Resource Locator
US United States
VPN Virtual Private Network
WIH Web Intelligence Hub
www World Wide Web

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 8
Abbreviations

 Table of contents

 Introduction

Introduction
Eurostat is the statistical office of the European Union. Its mission is to provide high quality statistics
and data on Europe. In order to reap the benefits of the data revolution and to complement traditional
statistical methods and sources with innovative approaches, Eurostat has recently established the
Trusted Smart Statistics initiative that will enable the development of IT, methodological and quality
frameworks, guidelines, tools and infrastructure that are suitable for big data processing. As part of
this Centre, Eurostat in collaboration with national statistical institutes (i.e. the European Statistical
System) will produce official statistics based on information that can be retrieved from the world wide
web (www). In the framework of the so-called Web Intelligence Hub, Eurostat has initiated
experimental research towards extending the capabilities of retrieving information on European
drone businesses from the internet.
The ‘Web Intelligence for Drones’ initiative builds on previous research on the web scraping of
business information in the context of official statistics. The study was triggered by the absence of
consolidated data on drones, the operation of drones or market size in the EU, despite the EU having
one of the most advanced regulations on unmanned aircrafts in the world. In particular, research
targets the development of a methodology and of the tools needed to retrieve information from the
www for businesses based in EU countries that have their main activity in the civil drones sector. The
methodology and tools are developed with a perspective of generalisation, i.e. it targets their
application to other emerging economic trends.
In a consolidated manner, this report presents the work performed to acquire web intelligence
concerning the drones sector, describing the methodologies developed and the results obtained.
Furthermore, it informs on the application of the methodologies developed to other emerging
economic trends.
This report is structured as follows:
  Chapter 1 introduces previous research initiatives on the web scraping of business information
 undertaken by the European Statistical System (ESS) partners, and it sets the legal background of
 the drone use case, with a particular focus on the European Drones Regulation.
  Chapter 2 presents the search approaches explored for identifying drone company websites
 through the www.
  Chapter 3 presents the method and scripts developed for the retrieval of URLs.
  Chapter 4 presents in detail the classification model and its extension to other countries studied.
  Chapter 5 presents the data extraction method. It informs about the analyses related to the
 activities of drone companies in the value chain and the analyses related to their location.
  Chapter 6 discusses the generalisation of the method and tools, and it presents an automated
 keyword extraction method.
  Chapter 7 presents the main findings and recommendations.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 9
Background 1

 1 Background

1. Background

1.1. Previous ESS research initiatives on web scraping of
 business information
The project builds upon previous research on the web scraping of business information in the context
of official statistics. It aims to further extend the capabilities of retrieving information on businesses
from the internet, in particular for the use-case of drone businesses, albeit with a view to generalising
the approach to new emerging economic trends. In particular, the research the project builds upon is
the work accomplished within the framework of the ESSnet Big Data I Work Package 2 (WP2) and
the ESSnet Big Data II Work Package C (WPC), which focused on enterprise characteristics.

ESSnet Big Data I WP2 was set up as a pilot study to investigate whether web scraping, text mining
and inference techniques can be used to collect, process, and improve general information about
enterprises. Specifically, it aimed to ascertain, whether business registers can be improved by using
web scraping techniques and by applying model-based approaches, so as to predict the values of
certain key variables for each enterprise, together with verifying the possibility of producing statistical
outputs using predicted data, in combination, or not, with other sources of data (survey or
administrative data).

Based on information gained from the enterprises’ websites, the specific use cases explored
included estimating, whether an enterprise performs e-commerce or not, whether an enterprise
manages job vacancies on its site, whether an enterprise is present in social media, its contact
information as well as other profiling information (e.g. type of business, links to other enterprises,
etc.). The use cases were implemented following a four-step process flow: internet access (i.e. URL
retrieval and website scraping), storage (organisation of unstructured scraped data), data preparation
(i.e. processing of texts), and analysis (i.e. machine learning or deterministic methods).

Two main categories of outcomes on enterprise characteristics produced by the ESSnet
Big Data I WP served as a starting point for continuing the initiative within the framework of the
ESSnet Big Data II WPC. These are: the methodology of URL (Uniform Resource Locator) retrieval,
process and software implementation, together with the methodologies, processes and software
implementations for detecting enterprises’ characteristics. These outcomes proved to be suitable for
implementation in all ESS countries, with some adaptations being made to local circumstances.

The ESSnet Big Data II WPC focused on the generalisation and extension of methodology and tools,
in line with ESS countries’ particularities. Implementing WPC involved the massive web scraping of

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 10
Background 1
company websites, the collecting, processing and analysing of unstructured data, and the
dissemination of national-level experimental statistics. Alongside publishing experimental statistics
(data and metadata) for the participating countries, the project also developed several policy
orientation and methodological guidelines, including an ESS-web scraping policy template (draft for
consultation with ESS partners), the Reference Methodological Framework for producing generalised
and extended methods, procedures and implementation requirements for web scraping on enterprise
characteristics, the Starter Kit for National Statistical Institutes, which consists of procedures for
testing and maintenance of web scraping, and the Quality Template for statistical outputs that are
based on web-scraped enterprise characteristics.

As a standalone initiative at country level, but also building on the ESSnet Big Data projects,
Statistics Netherlands (CBS) has developed a big data method for the web scraping of company
websites with the aim of identifying small innovative businesses that are not covered by the
innovation survey the institute carries out, whose population frame consists of companies with more
than 10 employees. Based on the content published on the websites’ main pages, the software tools
developed distinguish between innovative and non-innovative businesses. Classifying companies on
the basis of the text they publish on their websites proved to work well for innovation, as the survey-
based results could be replicated by the text-based approach developed. A copy of (nearly) all
business websites in the Netherlands was necessary to attain that goal. With small companies
however, dealing with the dynamics of the link between businesses, websites and the classification
of the websites of (semi-)self-employed is challenging. A model was recently developed to identify
platform economy websites that are active in the Netherlands.

Developing a method to automatically link URLs to the business register, more specifically the
domains of legal units, was the fruit of another standalone initiative by CBS in collaboration with the
private entity Data Provider. The method covers enterprises of all sizes although the process is more
challenging for recently established enterprises.

In all cases, the research projects listed above yielded successful results, and these form the basis
of the ‘Web Intelligence for Drones’ project. The recent study did however highlight specific additional
challenges that had not (yet) been addressed in the previous projects. The main statistical
challenges are: (given the absence of a classification of drone businesses as per the current NACE
Rev 2) the universe the study builds on the basis of the information existing drone operators’
websites provide, the lack of a website for some drone businesses (in particular if these are start-ups
or individual businesses), the lack of data (both at national and at EU level) that could serve as a
benchmarking reference to the estimates being calculated on the basis of data retrieved from the
web. The sector’s dynamic and innovative features, its companies’ rather small size and the
existence of borderline activities (e.g. drone operator vs drone business) presented additional
challenges that were explored within the scope of the project.

 The Web Intelligence for Drones project
  Measuring emerging economic trends through web sources, together with the particular
 use case of the drone industry, provide an input for future developments within the
 European Statistical System (ESS) Trusted Smart Statistics framework and the Web
 Intelligence Hub (WIH).
 In the context of the project, and more specifically in this report, the term Drone(s) is broadly used to refer to any Unmanned Aircraft
 Vehicle (UAV) or Unmanned Aircraft System, indistinctively.

 A UAV, commonly known as ‘drone’, is an aircraft without an on-board human pilot. UAVs are a component of a UAS that includes a
 UAV, a ground-based controller and a communications system between the two. There is a wide range of UAVs ranging from light
 and simple to heavy and complex aircraft, which operate with various degrees of autonomy and a diverse set of missions.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 11
Background 1
1.2. Legal background: the Drones Regulation in Europe
Since the beginning of 2019, the European Union Aviation Safety Agency (EASA) has been working
with the European Commission (EC) to provide common European rules that could improve the free
circulation of drones within the European Union (EU). A key point was to develop a common and
interoperable registration system that has been designed by the international public-private
partnership SESAR (Single European Sky ATM Research) Joint Undertaking. Pursuant to Regulation
EU 2019/947, this common framework registration became mandatory in January 2021, targeting the
conversion of all national authorisations, certificates and declarations to the new EU system by July
2021. Full operability of the common registration system may be expected for 2022 (7).

The certification of a UAS (Unmanned Aircraft System) implies collecting information on drone
manufacturers, given that, for each drone that is subject to certification, a unique digital registration
number will be issued. The information collected includes the manufacturer’s name, and it allows for
individual identification. On the other hand, UAS operators whose operations may present a risk to
safety, security, privacy, and protection of personal data, or to the environment, should also form part
of the EASA register. One must note that most operations do present the risks mentioned above, as
most UASs include a camera. Additional information is required for the mandatory registration of
operators whose activities present a risk, and this goes beyond basic information. The additional
information requested may include the activities the operators usually perform. Based on the
information available at national level, the main activities of some of the national authorities are
inspections/monitoring, flight tests, photography, pilots’ training, and others.

An important question is whether the lists of registered manufacturers and operators will be publicly
available in the future. The EASA registration system differentiates between natural and legal
persons; therefore, the first list could be brought under the protection of the General Data Protection
Regulation (GDPR). It is assumed that the second list (i.e. registered legal persons) could be made
available to other EC departments, if not made publicly available. Once the common European
registration system becomes operational, these lists’ future availability is important to the exploratory
sector analysis and to the validation of results collected from web data sources for the specific use
case of drone businesses.

1.3. Analysing the industry through company websites
An industry is composed by a group of companies that buy from suppliers of similar goods and
services to offer similar products and services to their buyers (Rothaermel, 2019). Official registers
organise companies according to national classifications of activities. In the EU, the NACE
classification allows to identify a company with the sector in which it performs its main activity. This
brings data to measure the importance of a sector and its impact in Europe. Companies usually
disclose this information on their websites, where they may also reveal their main activities as well as
the sectors they target.
Analysing an industry from the economic perspective indicates the number of businesses in that
industry, where they are located, the impact the industry has on its location and neighbouring
locations (in terms of GDP, of employment), and the impact an industry has on other sectors (Boix et
al., 2015; de Miguel Molina et al., 2012).
Over the last years, academic researchers have developed methodologies to extract information
from company websites that might well become a substitute for surveys, especially for the analysis of
emerging sectors such as nanotechnology, graphene and green goods (Youtie et al., 2012, Shapira
et al., 2014; Kinne and Axenbech, 2020). Statistical experts have also developed methods that might

(7) See for further details: https://www.easa.europa.eu/newsroom-and-events/news/drone-guidance-extended-and-
updated-support-safe-drone-operations-easa

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 12
Background 1
substitute national innovation surveys as they are carried out nowadays (Daas and Van Der Doef,
2020).
Previous studies cite the advantages in using websites as a source of information, such as public
availability, lower costs, complementary data, increased timeliness, and suitability to work with time-
series data through the Wayback Machine (Youtie et al., 2012; Shapira et al., 2016). They do
however indicate some challenges with using websites as a source of data, such as the difficulties in
accessing some of them due to their design (Arora et al., 2016; Shapira et al., 2016) and the fact that
the information disclosed differs, depending on the company, sector and country (Arora et al., 2013).
The studies analysed also indicate that keywords might capture various meanings and this would
imply searching for more information about a company (Shapira et al., 2014). Other authors find that
less information is available for big companies when they use the Wayback Machine (Gök et al.,
2015), which also entails a challenge when subdomains are growing due to new investors or new
languages, as the company enters new markets (Arora et al., 2016).
The studies analysed mainly focused on SMEs. Three main tasks can be detected in the methods
they performed: a) searching for a list of companies in a different way than through using activity
codes, b) extracting the necessary keywords from websites, and c) analysing the information to
classify websites based on the keywords they include. The main characteristics of previous studies
exploring company websites as a source of information for the analyses of industries are listed in the
following table.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 13
Background 1
 Table 1: Main characteristics of previous studies exploring company websites as a source of information for industry analyses
 Use
 Sample information How information is
 Reference Sector analysed List of companies How information is extracted
 (companies) from Wayback analysed
 Machine
 Ellinger et al. (2003) Motor carrier Top 100 motor carriers 98 No Content analysis Descriptive
 Libaers, Hicks and A. Search for 89 keywords on firms’ Factor analysis with
 Various ≤ 500 employees & ≥ 15 patents 407 US SMEs No
 Porter (2010) websites keywords
 Youtie et al. (2012) Nanotechnologies List available 30 US SMEs Yes Web scraping, content analysis Cluster analysis
 Search firms online, in papers, 20 SMEs in 3 Web scraping, searching for
 Arora et al. (2013) Graphene Yes Cluster analysis
 patents and business database countries keywords on webpages
 Green goods Search companies (and their
 (green technology employees) in Fame database, with a 304 UK Search for list of keywords on
 Shapira et al. (2014) No Descriptive
 manufacturing list of keywords (from different companies companies’ websites
 companies) sources)
 296 SMEs UK-
 Database by Shapira & Harding Web crawling and text mining Descriptive and
 Gök et al. (2015) Green goods based Yes
 (2012) and Fame database (237 out of 296 had a website) correlations
 companies
 300 SMEs US-
 Arora et al. (2016) Green goods Own database from a project based Yes Web crawling ---
 companies
 65 SMEs
 Search companies in patents, social Cluster analysis and
 worldwide Web crawling and web mining
 Shapira et al. (2016) Graphene media, papers, business databases, No binary logistic
 (manufacturers (content analysis)
 reports, websites regression
 )
 Green goods Search in D&B database with 100 Web crawling and content Hausman–Taylor
 Li et al. (2018) 300 SMEs US Yes
 manufacturers keywords analysis estimation
 Search in D&B database with 100 Web crawling and content
 Arora et al. (2020) Green goods 223 SMEs US Yes 2SLS regression
 keywords analysis
 Categorization/
 List from the Community Survey on 4755 Italian
 Bruni & Bianchi (2020) Various No Web scraping and text mining classification
 ICT usage and e-commerce companies
 algorithms
 Classification
 Daas & Van Der Doef 824,972 Dutch
 Various Companies in CIS Netherlands No Web scraping algorithms, logistic
 (2020) companies
 regression
 Web scraping (ARGUS,
 2.4 million
 Kinne & Axenbech proprietary software) and data Descriptive and probit
 Various Mannheim Enterprise Panel German No
 (2020) mining (1.15 out of 2.4 million regression
 companies
 firms have URL)
 912,850 Web scraping (with ARGUS), web
 Descriptive and probit
 Mirtsch et al. (2021) Various Mannheim Enterprise Panel German No content mining and web structure
 regression
 companies mining
Source: own elaboration based on research published in the Web of Science database

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 14
Background 1
When one considers the papers analysed in Table 1, the main challenge to the exploratory sector
analysis is that of obtaining a list of company websites. Most of the industry analyses based on
company website information (see Table 1) made use of an already existing database. The
advantage of those datasets is that they offer additional information for further analyses, such as the
number of employees in each company, for example. When such a list was not available, the authors
created a list of keywords and searched for them in a commercial database (Fame) (8) or in the Dun
and Bradstreet (D&B) database. Alternative solutions available were: i) buying access to a list of
companies and their websites or ii) directly searching on the www. In all studies that directly
searched on the www, the search was performed manually and it only made use of a single search
engine.
Developing a method to create lists of drone businesses from the information that is available on the
www is explored through the search strategies described in Chapters 2 and 3.

1.4. Exploratory analyses of the drone sector
Exploratory drone sector analyses were performed at country level, in a number of Member States
(Spain and Ireland). This provides a more thorough conceptual understanding of the sector and its
characteristics. It also acts as an input into the development of methodologies and tools for the
retrieval of information on the sector from the web. A web search strategy was developed based on
the exploratory sector analyses, with the aim of identifying the universe/population of drone
businesses, based on web content, and to explore the feasibility of collecting information on these.
The main findings of the exploratory drone sector analyses in the two countries are provided in the
following paragraphs.
Although the universe of the study is built on the existing websites of drone businesses, for the
purpose of analysing the industry, a starting point taken was to identify publicly available lists of
drone operators and businesses. This facilitates the identification of drone-business websites and, in
particular, it furthers the exploration of the information one can extract from the websites
(i.e. classification and variables).
In both countries for which drone sector exploratory analyses were carried out, the study team made
use of publicly available information concerning registered drone operators or companies. The
sources of information were: for Spain, AESA (Agencia Estatal de Seguridad Aérea) and, for Ireland,
IAA (Irish Aviation Authority), UAAI (Unmanned Aircraft Association of Ireland) and the Drones
Ireland Directory website. Several lists of drone operators and companies were compiled and
analysed from these web sources.
For Spain, the AESA website yielded the following registers:
  A list of approximately 5 200 drone operators (as of December 2020), including individuals
 and companies, and details on the activities these were authorised to undertake;
  A list of approximately 67 companies (as of December 2020), including information on
 specific operational scenarios (9). Those 67 companies were also all listed in the list of drone
 operators (see first bullet point);
  A list identifying 11 Spanish drone manufacturers.
For Ireland, the three web sources analysed yielded the following registers:
  A list of 350 drone operators authorised by the IAA, including the names of professionals,
 companies and offices of local governments registered through the IAA;
  A list of 5 flight schools and 65 UAAI members. 42 % of the records in this list overlapped

(8) https://fame.bvdinfo.com/
(9) Three application scenarios are defined in the new EU Drone Regulation: Open, Specific and Certified. They depend
exclusively on the flight‘s risk level, independently of the nature of the use of civil drones (leisure or professional).

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 15
Background 1
 with the IAA list (see first bullet point);
  A directory of drone professionals and companies operating in Ireland, which allowed for the
 identification of those entities having their headquarters in Ireland. Out of the 56 results
 obtained for this directory, applying the headquarter filter, 29 % of records overlapped with
 those in the IAA list (see first bullet point).
In Spain, one could distinguish a certain level of centralisation and consistency, in the information on
drone operators and companies. All three lists were published by the national air safety authority
(AESA in Spain) and the relevant records in the shorter lists were, in all cases, already recorded in
the comprehensive list of all drone operators authorised in Spain. The shorter lists published by the
same source (i.e. AESA) were helpful for the further characterisation of specific companies. In
Ireland, the additional websites with information on drone companies and operators complemented
the list published by the IAA, not only with additional information, but by also extending the list of
known companies or operators.
Based on these web sources listing operators and companies that are active in the drone sector in
the two countries, a sample of companies was selected. For Spain, the sample was given by the
companies listed in the short list (i.e. the 67 companies authorised for specific operational scenarios
and the 11 Spanish drone manufacturers). The final sample thus included 76 companies (two
companies were referred to in both lists). For Ireland, the sample was selected based on the IAA list.
Namely, those operators that are located in Dublin County were selected (a total of 110 operators,
roughly 1/3 of the total list for the whole country).
For both countries, the URLs of selected companies were picked manually. In Spain, a website was
available for each company in the sample while, in Ireland, no website could be found for 13 of the
110 companies.
In order to analyse the information extracted from the identified websites, two steps were followed: i)
web observation and definition of variables, and ii) content analysis with the support of QDA Miner 5.
Step 1 also led to the identification of individual companies, specifically in the case of Ireland. The
detection of those, along with other exclusion criteria (i.e. companies not headquartered in Ireland)
further reduced the sample of companies analysed. The final sample size in Ireland amounted to 82
companies. Furthermore, web observation was used to obtain data on a set of predefined variables
(e.g. availability of information about drones on the specific website, the economic sector the
company is active in, its main activity, value chain identifiers). Step 2 saw the identification of key
words that are used by companies to refer to ‘drones’, the type of services or operations the
companies offer, with or in relation to drones, and the economic sectors targeted by the company’s
offer).
The drone industry exploratory analysis showed that many company websites do not include any
information concerning drones. In the case of Ireland, only 44 % of the companies do disclose
information about drones on their webpages.
The industry analysis showed that, in both countries, the sector presents a different structure,
influenced, one could say, by creative industries in Ireland and by engineering in Spain. In other
words, the information disclosed by companies differs between countries, which could lead to errors
if the quality of the websites’ content reflect sectoral reality.
The data obtained through the industry analysis indicate that:
  The sector’s value chain can be represented, based on the following activities:
 1. Services,
 2. Manufacturer,
 3. Design,
 4. Components,
 5. Software,

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 16
Background 1
 6. Training,
 7. Consultancy,
 8. Distribution,
 9. Rental of equipment.
  The services drone businesses offer to other companies and customers can be identified,
 and the following were identified as being the most common:
 1. Inspections,
 2. Mapping,
 3. Aerial photography,
 4. Aerial video,
 5. Surveying.
  The sectors drone businesses target with their offer can be identified. A classification could,
 for example, include:
 1. Agriculture,
 2. Infrastructure,
 3. Architecture,
 4. Real estate,
 5. Construction,
 6. Television, etc.
  Identifying the services offered and the sectors targeted by the drone businesses allows one
 to establish a link to those sectors’ NACE codes and thus to classify the companies
 according to those codes.
  It is possible to ascertain drone client companies’ specialisations as well as their
 geographical area / location.
  Economic variables, such as the number of employees or turnover, are not usually available
 on the websites.
The exploratory analyses performed were essential to defining the keywords that are to be used by
the web search strategy (see Chapter 2), to the classification model applied to drone companies (see
Chapter 4), and to data extraction (see Chapter 5). Search engines do not perform well if too many
different words are used in a query. The selection of search terms should therefore be based on the
preliminary analysis on a small sample. Keywords will also be needed in the classifying phase, given
that those terms will be included in scripts to search for them on the websites.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 17
Approaches for searching for companies websites 2

 2 Approaches for searching
 for companies websites

2. Approaches to searching for company
 websites
Searching for the websites of companies that are active in a specific sector and in a particular
country can be accomplished in several ways. The approaches explored for this study are discussed
in this section. But first, some introductory comments and general remarks are provided.
When using a search engine such as Google, Bing or DuckDuckGo, it was determined that the
findings can be affected by:
  the location (IP address) of the searching user,
  the user’s previous search history (cookies),
  the country extension of the search engine used (e.g. using Google.nl vs. Google.ie), and
  the User-agent of the ‘browser’ program used.
These effects can be reduced (although not entirely) through:
  using a VPN connection,
  using a browser that has no search history or searching the web via an anonymised
 (incognito) browser, and
  using a search engine that is specific to the country under study.
For the initial studies described below, a VPN connection with a Luxembourg (10) location and a
generic Mozilla browser User-agent identifier were used to mimic, as closely as possible, the search
results of a browser program used at Eurostat’s premises. All queries were coded in Python
Version 3.
The following search strategies were identified and investigated.

(10) For subsequent tasks the program’s VPN connection was set with a location in Belgium.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 18
Approaches for searching for companies websites 2
2.1. Searching for websites that provide an overview of
 drone companies

The first step of this approach involves choosing several drone-specific terms, a term for ‘list’,
‘overview’ or ‘pdf’, and the country’s name. For countries, in which the main spoken language was
not English, two strategies were used. The first made use of search terms in the main spoken
language including the word drone, drone abbreviations, and other words specific to several
activities. The second approach used the same words but all written in English. The first 50 links
provided by the search engines were checked and any links referring to websites with country codes
other than the country under investigation were removed. For example, when studying Irish websites
(such as www.drone.ie) any link referring to a UK website (such as www.dronedirect.co.uk) was
removed. Please be aware that any overview of drone companies may be included in webpages
(html-files), pdf files or both. In both cases, it subsequently becomes necessary (where available) to
extract the references to websites. Other than references to websites, the names of companies were
also identified and extracted. This step was followed by a search for the URLs corresponding to the
company names obtained.

 Findings:
  ‘Pros’: This is a very efficient strategy from a search-engine perspective. Only one search
 query is needed. It can work very well on specific countries, in particular when national air
 safety authorities publish overview lists of drone operators (e.g. Spain).
  ‘Cons’: This strategy does not work for all countries. For example, no such overview
 website could be found for Germany and Italy, indicating that this strategy may never be
 the only approach to follow.

2.2. Searching for websites of individual drone companies
This approach begins with the selection of several terms that are specific to drones, as well as the
country's name. For countries, in which English is not the main language, two search strategies were
applied. In the first, the search terms used, except for the drone abbreviations, were all written in the
country’s main spoken language. In the second strategy, all words were written in English. All links
found by the search engines (up to between 200 and 300) were checked, and any links referring to
websites with country codes other than the country being investigated were removed.

 Findings:
  ‘Pros’: This strategy enables one to find websites of individual companies. This may
 include those that are missing the from overview websites.
  ‘Cons’: This strategy requires considerable amounts of search requests to be performed,
 which calls for the necessity to regulate requests over time, so as to prevent the search
 engine used from blocking. At the time of drafting, it is unclear why only up to 300 websites
 could be found.

2.3. Finding drone websites via pictures of drones
This approach begins by searching for pictures of drones in a specific country. This is done by
limiting the search query to image-type results. Next, the domain names of the links of the images
found are collected. The results were not satisfactory.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 19
Approaches for searching for companies websites 2
 Findings:
  ‘Pros’: Using this strategy, websites are found that contain pictures of drones.
  ‘Cons’: Because of the image-oriented approach, considerable amounts of websites are
 detected, on which drone-enthusiasts post pictures. These are however often blog-like
 websites and the websites of other drone fora. Compared to strategies 2.1 and 2.2, a fairly
 limited number of drone company websites were found.

2.4. Finding drone websites via a company’s impressum
 page
This approach focuses on finding the ‘impressum’ pages of companies containing drone-specific
words in a given country. This is especially interesting for Germany, for example, where many
companies are legally required to have an impressum page on their website. Such pages provide
legal and public information to customers, in a standardised way. Basically, a search query is
performed including drone-specific words, the country’s name and the word ‘impressum’.

 Findings:
  ‘Pros’: Using this strategy, websites of drone companies are found to include drone-
 specific words on their impressum webpage.
  ‘Cons’: Depending on the country, only a limited number of companies may have an
 impressum page. The page in question must also include a drone-specific word. Due to
 that, a limited number of companies are usually identified, compared to search strategies
 2.1 and 2.2.

2.5. Finding drone websites using a non-search engine
 approach
Apart from using a search-engine based approach, one may also attempt to find drone companies in
another way. For example, one can search for open access Chamber of Commerce data as well as
for websites that provide overviews of the companies that are active in a specific region of a country.
Using drone-specific words as search terms, one may find the names of companies active in the
drone industry, when such words form part of the company name or if they are included in the short
description (if provided).

 Findings:
  ‘Pros’: Compared to the other strategies, this is – currently – the only way in which to
 detect drone companies without a website.
  ‘Cons’: The websites searched may not include URLs or they may provide a limited
 amount of URLs for the companies listed. This raises the additional need for a search
 strategy to find those companies’ URLs (if available). In the ESS, this constraint it tackled
 through the search strategy that was already developed in the context of the ESSnet Big
 Data II Workpackage C – Enterprise characteristics.

In addition to the findings listed above, a number of additional observations were made during the
study’s inception phase.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 20
Approaches for searching for companies websites 2
The first is the need for extracting URLs from documents including websites/html files, pdf-files, etc.
Specific approaches need to be developed for each type of file.
The second is the need to identify and extract company names from documents providing an
overview of drone companies, so as to determine those companies’ URLs (if available). This often
resulted in finding copious amounts of names, including those of many non-drone companies, whose
URLs subsequently needed to be searched for.
Thirdly, it is necessary to develop a method able to determine, whether an identified website really
belongs to a company active in the drone industry. Because the search strategy is applied in multiple
countries, in which different languages are spoken, it is necessary to develop a word-based set of
rules to discern between drone and non-drone companies. The aim is to have a specific set of words
available to use to identify a drone website in other European countries. This approach was
implemented at the end of the search strategy, as it was found that, toward the beginning of the
search strategy, it is much more efficient only to remove URLs that clearly do not correspond to
companies active in the country under study.
Lastly, an intriguing question arises: how exactly does using a search engine determine, whether a
website can be found or not? Imagine a small drone company that has only recently created a
website which, so far, has not attracted any visitors. It is unlikely that such a website would be
covered by any of the strategies making use of a search engine. The latter must be correct as a
website’s popularity is the result of the number of links to that website found by the various search
engines. Because companies must register with their country's Chamber of Commerce, it is
extremely likely that the type of 'young' company described may only be detected by the fifth search
strategy. The same holds for companies that do not have a website.

 Web Intelligence for Measuring Emerging Economic Trends: the Drone Industry 21
You can also read